[BioC] Remove batch effects from RNA-seq data using edgeR and sva/ComBat

Christopher Conley cjconley at ucdavis.edu
Tue Sep 10 20:16:36 CEST 2013


Having had personal communication with Dr. Evan Johnson 
on this very question, I am quoting his email response. 

> So my question is: Are there any reasons why using ComBat 
with RNA-seq data is not legit? 

Here is what Evan had to say:

"For batch effects, if the sample sizes are large, say around 
10 per batch or more, ComBat and SVA will work fine 
regardless of whether they are on count data or not. For 
cases with 50-100 per batch, ComBat and SVA will work 
extremely well and will be somewhat optimal. Basically this is 
due to the Central Limit Theorem. For small batch size, 
ComBat and SVA are still valid, but there may be some 
research that can be done here."

Hope that helps,
Christopher Conley
Graduate Group of Biostatistics
UC DAVIS



More information about the Bioconductor mailing list